OpenAI has officially introduced Presence, a platform for deploying AI agents in corporate environments. The solution is focused on automating workflows and customer interactions, offering companies a tool for creating autonomous assistants capable of performing tasks without constant human involvement.

Unlike traditional chatbots, Presence uses advanced language models integrated with access control systems and security rules. Agents can connect to internal databases, perform authorized actions in corporate systems, and transfer the conversation to an operator when human intervention is required. This makes the platform a flexible solution for tasks ranging from processing returns to insurance claims.

Key Features of Presence

Among the stated capabilities is access control, allowing the agent to only obtain the data and permissions necessary for a specific task. The Codex system automatically analyzes gaps in conversations and suggests algorithm updates, ensuring continuous performance improvement. Multimodality is also supported: agents can function both in text chats and real-time voice channels.

OpenAI has already implemented Presence in its English-language telephone support service. According to the company, the AI independently resolves 75% of incoming requests. The use of Codex reduced the frequency of call transfers to a human by 15% in just 10 days—an impressive figure that demonstrates the platform's scalability potential.

Among the first testers of the technology are BBVA bank, telecommunications giant SoftBank, and insurance group IAG. These companies are already using Presence to automate banking operations and support customers during peak load periods. At the launch stage, the product is available to a limited number of corporate clients, with implementation handled by OpenAI engineers and certified partners.

My analysis. Presence is a logical step by OpenAI into the B2B segment, where demand for autonomous AI solutions is growing exponentially. However, the key challenge remains the balance between automation and control: even with 75% successful requests, 25% require human intervention, which in critical industries such as finance or insurance may be unacceptable. Nevertheless, integration with Codex and rapid results in reducing operator workload indicate that the technology is mature for mass adoption within the next 12 to 18 months.